The present invention relates to the measurement of features that are patterned by photolithographic techniques, and more particularly to such measurement using scatterometry.
A variety of articles such as integrated circuits and microelectromechanical machines have micro-scale and nano-scale structures that are patterned on a substrate or wafer by photolithographic techniques. Such structures have critical dimensions ranging in size from tens of nanometers to a few hundred nanometers and single microns. Critical dimensions are structural details in a product that have been identified as key to monitoring and controlling the manufacturing process. The structural detail is required to be within a certain range or tolerance of a design size. Failure of the structural detail measurement to be within tolerance of the design size initiates corrective actions such as rework and process control parameter change.
When fabricating such nano- and micro-scale structures, measurements of critical dimensions must be taken at several stages during fabrication. In this way, the results of prior processing are measured and checked against tolerances to determine if acceptable. If the results of prior processing are not acceptable, the wafer is removed from the manufacturing line, before additional process steps are performed and costs incurred. Actions are then applied to the manufacturing line to correct process problems.
An existing metrology technique for performing such measurements is known as critical dimension scanning electron microscopy (“CDSEM”). CDSEM is a dominant metrology technique used in semiconductor and other nano-scale manufacturing. In CDSEM, a beam of electrons having a spot size of a few nanometers (typically five to 10 nm) is scanned in a raster pattern across the surface of a sample to be imaged. The electrons of the beam striking the surface cause secondary electrons to be given off from the sample. A detector of the scanning electron microscope picks up a signal representative of the secondary electrons, which is then processed to drive a display which is raster scanned in unison with the electron beam. The resolution of CDSEM depends upon a number of factors including the spot size of the electron beam on the sample.
CDSEM is an expensive measurement technique. CDSEM is expensive because electron microscopes require electron sources and focusing elements that are orders of magnitude more expensive than optical systems that use light. It also has insufficient throughput to operate on more than a handful of wafers that are pulled from the manufacturing line as samples representative of the manufacturing process.
A disadvantage of CDSEM is that only features on the surface of a sample can be imaged because the electron beam does not penetrate beneath the surface. When features below the surface of a sample are to be measured, other instruments must be used. CDSEM is one of a class of instruments known as a reference measurement system (RMS). Other RMS instruments include CDAFM, cross section SEM, and other metrology instruments. CDAFM is an atomic force microscope designed to make critical dimension measurements. Cross section SEM is a specially designed scanning electron microscope system used to image and measure a patterned wafer in cross section. When a buried structure of a wafer is measured by cross section SEM, the wafer is cleaved, i.e., intentionally fractured along a lattice boundary of the semiconductor crystal. The cleft exposes critical dimension features for measurement, and an electrographic image is taken of the feature. Cross section SEM operates destructively. After the wafer is cleaved, it cannot be processed further in the manufacturing line, and thus the wafer is rendered unusable.
The speed of a metrology system is generally defined in terms of move acquire measure (MAM) time. The MAM time includes: a) the time required to move the article to be measured from one measurement location to another; b) to locate the new site to be measured (which may necessitate acquiring an image for pattern recognition); and c) to take the measurement. CDSEM provides a longer than desirable MAM time. CDSEM requires the sample to be placed in a high vacuum with the CDSEM tool. Therefore, each time a wafer is imaged, the tool must be brought to atmospheric pressure to permit the wafer to be loaded into a chamber housing the tool and wafer, and the chamber must then be pumped to a high vacuum, typically 10−7 torr or less. In addition, the constraints of electron beam imaging force CDSEM measurements to be taken over a relatively small area of the wafer. Because of that, the measurement from just one location of a wafer may not be generally representative of prior processing. Therefore, CDSEM tools typically take several measurements from different locations of the wafer, e.g. five locations, and take the average of the measurements as being representative of prior processing. Thus, the MAM time per wafer for a CDSEM tool requires load and unload time for placing the wafer in the vacuum with the tool, and the time for taking the measurements at the required number of different locations (e.g. five) of the wafer.
CDSEM provides an appropriate basis for comparing the accuracy and precision of other measurement systems, as well as cost-of-ownership (“COO”). Cost of ownership (“COO”) is defined as the total cost of using an instrument or process in manufacturing. Key elements in determining the COO are equipment purchase and maintenance cost, and throughput in using the equipment. For metrology instruments, throughput depends on several factors including the MAM time, and the number of locations of the wafer to be measured. CDSEM, while providing adequate measurement accuracy and precision, has less than desirable COO. As mentioned above, electron beam imaging systems are very costly to purchase and maintain. In addition, throughput is undesirably low, for the reasons discussed above making the cost per wafer high for using the tool.
Scatterometry now provides an acceptable alternative to CDSEM for measuring micro-scale and nano-scale features that are formed by photolithographic techniques. Scatterometry is an optical measurement technique that is both nondestructive and which can operate in-line, without requiring wafers to be taken out of the manufacturing line. Scatterometry operates by measuring the characteristics of light or other radiation that diffracts or “scatters” off a grating, the grating being representative of features of the wafer to be measured. Such grating, also called a “target”, typically has lines of dimensions comparable to a critical dimension of the features.
Until recently, the calculations required by scatterometers to analyze the return signal in relation to the critical dimension were too computationally intensive to perform with acceptable MAM time. With recent advances in computing, that situation has now changed, and scatterometers are now available which provide acceptable MAM time.
Scatterometers differ in light sources, scattering conditions, and method for analyzing the return signal coming back from the target. Scatterometers are available for use with simple line grating targets to determine critical dimensions such as structure bottom width, structure sidewall angle, and structure height. Given future advances in computing, it is expected that scatterometry will be able to analyze light scattered off of arrays of more complicated structures such as contact holes and in-chip periodic structures.
Scatterometry offers several advantages over CDSEM for measurement of critical dimension features. Scatterometry measurements are performed at atmospheric pressure. Thus, scatterometry avoids the load and unload time of CDSEM tools for placing a wafer in a high vacuum for taking measurements. In addition, because of the large size of the targets used in scatterometry, a single measurement inherently averages out line edge roughness and other random variations in the grating. As discussed above, in CDSEM several measurements must be performed to average out roughness in the sample. In addition, the capital cost of a scatterometer is generally lower than a CDSEM tool. These advantages allow scatterometry to provide increased throughput and lower COO relative to CDSEM.
Unlike CDSEM, scatterometry techniques can detect materials and structures buried beneath the wafer surface. Another future application may be to use scatterometry for overlay metrology.
Scatterometry provides other advantages. It is expected that new critical dimension metrics such as sidewall angle and structure height will become necessary. Such metrics can help to better measure printing and etching processes in nano-scale manufacturing, which processes are plagued by loss of fidelity between the intended feature shapes and those that actually result. Scatterometry provides these new metrics. CDSEM does not provide such metrics.
Another advantage of scatterometry systems is that they can be integrated into process equipment used for etching and lithography. Some such process equipment targets a throughput of 120 wafers per hour. Because of the long MAM time of CDSEM tools that require measurements to be performed in a high vacuum, a CDSEM tool cannot meet this performance objective. Accordingly, only optical measurement tools are considered in connection with such high throughput process equipment.
A good scatterometry target mimics features of the operational area of the sample, having elements that vary in proportion to the variation in the sample features due to the manufacturing process that is performed. A grating formed in a layer of an article to be measured, having lines of the same critical dimension as the features of the layer, provides a good scatterometry target. With such grating, variations in the manufacturing process cause the grating to change in the same way as the critical dimension features.
Scatterometry targets must be at least a certain size in order to provide a return signal having adequate signal-to-noise ratio. This is due to the following. Critical dimension features are typically smaller than the wavelength of light used in scatterometry. In manufacturing semiconductor devices, lithography is used to define critically dimensioned features at nominal widths of less than 100 nm. On the other hand, available scatterometers use light having wavelengths greater than 200 nm. The situation is even more challenging than inferred by these numbers since the precision and accuracy of the critical dimension measurement must be kept within a small fraction of the nominal measurement, e.g. to within about 2%. To achieve this level of measurement quality, the light must be scattered from many lines in the grating, for example 40 lines or more. The minimum number of lines determines the minimum acceptable grating size, and a minimum spot size for the light beam, as well. Additional constraints on the minimum grating size are imposed relating to the accuracy of directing the beam onto the grating.
On the other hand, the maximum size of the scatterometry target is limited by the area available for such target on the wafer, as the scatterometry target must compete for wafer area with other targets used for other types of metrology systems. In summary, while there is incentive to make the grating of a scatterometry target as small as possible, the requirements for precision impose a minimum size. Today, no commercially available scatterometer can precisely measure key parameters of critical dimension features with a grating smaller than 50 μm by 50 μm.
An example of a grating used as a conventional scatterometry target 10 is illustrated in FIG. 1. The grating 10 consists of parallel-oriented lines 11, each having the same width 15 as the width of critical dimension lines in an operational area of a wafer, and a spacing 14 which is the same as the spacing between the critical dimension lines. In such target, the dimensions 12, 13 of the grating are about 50 μm on a side.
Some types of prior art scatterometry systems are illustrated in FIGS. 2A through 2C. Types of conventional scatterometry systems include normal incidence spectroscopic reflectometry (FIG. 2A), spectroscopic ellipsometry (FIG. 2B), and two-theta fixed-wavelength ellipsometry (FIG. 2C).
In a normal incidence spectroscopic reflectometry system as shown in FIG. 2A, a broadband source 213 produces white light, having wavelengths between 200 nm and 800 nm. The light is focused by a lens system 214 and passed through a beam splitter 215 as a spot onto the target 212 of the sample 211. The return signal scattered off of the sample is then reflected by the beam splitter 215 and passed through further optics 216 onto a detector 217. The detector uses a prism or grating to separate the return signal into its constituent wavelengths. As shown at 218, reflectivity is then determined as a function of wavelength from the separated light.
In a spectroscopic ellipsometry system as shown in FIG. 2B, the light from a white light source 223, having wavelengths between 200 nm and 800 nm, is focused by a lens 224 onto a grating 222 of a sample 221. The light is reflected off of the grating 222 at a fixed angle of incidence and focused through a lens 225 onto a detector 226 having a prism or grating for separating the scattered light into its constituent wavelengths. The zeroth order diffracted light is then detected. The zeroth order light is that which scatters off at an angle that is equal to the angle of incidence. Rotating polarizers 228 and 229 are provided in the incident beam and the scattered beam of the return signal, respectively. As shown at 227, changes in the degree to which the returned light has transverse electric (TE) and transverse magnetic (TM) polarization are recorded as a function of wavelength.
With a two-theta fixed-wavelength ellipsometry system, as shown in FIG. 2C, a single wavelength of light from a source such as a helium-neon laser 233 is focused by a lens 234 onto a grating 232 of a sample 231. As in the system shown in FIG. 2B, the light is reflected off of the grating 232 at a fixed angle of incidence and focused through a lens 235 onto a detector 236 having a prism or grating for separating the return signal into its constituent wavelengths. As in that system, the zeroth order diffracted light is detected as a function of the angle of incidence. Again, rotating polarizers 239 and 240 are provided in the incident beam and the scattered beam of the return signal, respectively. During the course of a measurement the angle of incidence 238 and the scattered angle are kept equal and swept through a range of angles. As shown at 237, changes in the degree to which the returned light has transverse electric (TE) and transverse magnetic (TM) polarization are recorded as a function of the angle of incidence.
In principle, when the geometry of the grating and the optical properties of the materials involved are known, the scattering properties of the electromagnetic radiation incident upon the sample can be determined by solving Maxwell's equations. That is, the return signal from the sample varies in certain expected ways. From the known geometry and optical properties, properties including the variation in reflectivity with wavelength, variation in polarization with wavelength, variation in polarization with angle of incidence can be determined.
In general, however, the inverse problem cannot be solved. That is, the geometry of the grating generally cannot be determined, even with knowledge of the scattering properties and the optical properties of the materials.
With the inability to solve for the geometry of the grating, scatterometry systems rely instead upon the correlation of return signal characteristics with return signals obtained from simulations of samples having known characteristics. Such techniques operate as follows. An initial guess is made concerning the geometry of the grating to be measured. The scattering properties of the grating are measured, by which a return signal is measured in terms of spectra. The difference between the calculated and measured spectra is then determined. The difference is used to make a better guess as to the actual geometry. To decide whether the new guess is better than the first, a Chi-square sum of least squares quantity is calculated. The Chi-square quantity is the sum of squares of all the differences in spectra between the return signal and that calculated signal, over all wavelengths or angles of incidence. The smaller the Chi-square quantity, the closer the fit is between the measured return signal and the calculated signal.
One problem with the Chi-square sum of least squares approach is the local minima problem. The Chi-square quantity is a function of many parameters, all of which are allowed to vary in the geometry model. Thus, varying parameters of the model to search for the true minimum Chi-square value could lead to a local minimum in the multi-parameter Chi-square surface, rather than the true minimum corresponding to the true geometry.
A common way of overcoming the local minima problem to determine the true geometry is to search the whole parameter space in fine steps using a library based approach. With experience and some supporting metrology, a model of the grating is determined which includes those dimensions that can vary when the instrument is used.
FIG. 3 illustrates an example of a model for measuring such grating geometry. The model represents a sample to be measured having recurring resist patterns 30 of trapezoidal cross-section. The resist patterns are disposed over an unpatterned anti-reflective coating (ARC) layer 31, which in turn is disposed over other unpatterned underlayers 32 through 37 of various materials and thicknesses. Some of these thicknesses may need to be variable (floating) in the model description from which a library of spectra signatures is derived. It is known from experience that the height 25, sidewall angle 26, and bottom width 27 of the recurring trapezoidal patterns can change depending on lithography process conditions. Therefore, these properties are allowed to vary. Accordingly, the scattering spectra must be calculated for every possible value of these properties. Assuming that the bottom width 27 can vary between any value from 21 nm to 100 nm in one nm steps (chosen because of our accuracy and precision requirements), then there is a total of 80 different values for bottom width. Similarly, the sidewall angle 26 can be allowed to vary in one degree steps between 85 and 94 degrees for a total of 10 different angle values. In addition, the height 25 is allowed to vary in one nm steps between 151 nm to 250 nm for a total of 100 different height values. All combinations are calculated for a total of 80,000 spectral signatures. This is while assuming that the thickness of any underlayer remains constant, when in fact it may vary, and affect the measurements that are made. Thus, if underlayer variation is to be considered, an even greater number of spectral signatures must be calculated. In such manner, the spectral signatures of a geometric model library are determined and stored. At measurement time, the measured spectra are compared to the libraried spectra by a Chi-square approach to determine the best fit. The geometric properties of the model having the smallest Chi-square quantity are then selected as the best fit.
A variation on the library-based solver is to use a coarse library having large step sizes for the floating parameters to determine an approximate solution. This solution is assumed to be near the true minimum on the Chi-square surface. Therefore, mathematical techniques can be employed to “linearize” the problem and regress to the true minimum. In scatterometry systems, both techniques are desirably used to reduce the quantity of calculations that are required.
Unfortunately, the relatively large size of the scatterometry grating (50 μm) introduces additional lithographic printing problems or constraints, compared to the printing of the critical dimension element of the chip, which has a size typically smaller than one μm. Pattern collapse is a condition in which photoresist lines fall over, for example. Pattern collapse can occur in such large targets, making them unusable, even when the in-chip element has printed acceptably. Pattern collapse radically changes the geometry in unpredictable ways, making scatterometry results unreliable.
Therefore, it is highly desirable for the lines of the grating to remain standing whenever the printed element of the chip is standing so that scatterometry measurements track the properties of the printed element. In other words, the grating should be as robust as or better than the critical dimension printed element.
Pattern collapse is not unique to targets used in scatterometry. There are other situations where pattern collapse occurs. In the development of lithography processes, it is frequently necessary to cross section wafers by cleaving. Because the cleave crack can wander, target patterns have been designed into test reticles having arrays of long lines to improve the chance that the cleave crack will run through the lines.
As shown in FIGS. 4 and 5, to help avoid pattern collapse, some target patterns include bridges 20 (FIG. 4) or gaps 22 (FIG. 5). As long as the cleave crack runs through the lines 21 and not the bridges 20 or gaps 22, a suitable cross section is obtained. However, if the cleave crack intercepts the bridge or travels along the gaps, the cross section fails. For this reason these alterations of the simple line array are placed sparingly to make such interceptions improbable. Sometimes the alterations are placed along non-orthogonal directions, as shown in FIG. 5, so that even if the cleave crack intercepts one of the alterations, there will be other valid structures to measure. For taking the cross section, as long as the alterations do not interfere with the cross section measurement site, there is no measurement impact from these alterations.
While such alterations are suitable for patterns used to measure cleaved cross section sites, such as in scanning electron microscopy (SEM), it is quite another thing to apply such bridges and gaps to scatterometry gratings. Given the nature of the scatterometry measurement where scattered light from all structures within the probe beam spot contributes to the measurement, these pattern reinforcements would be expected to alter the spectra and also the measurement derived therefrom. In addition, the state of scatterometry modeling today only permits simple gratings to be modeled. Accordingly, heretofore, there has been no way to include such alterations in a scatterometry target in a way which permits a model to accurately represent them.
Accordingly, it would be desirable to provide a scatterometry target that mimics the behavior of critical dimensions of in-chip circuit elements, while having patterns that avoid collapsing. Such scatterometry target must also have good scattering properties for scatterometry measurement and analysis.
It would further be desirable to provide a scatterometry target having features that change more than the critical dimension features of the in-chip circuit elements due to manufacturing process variation. With such scatterometry target, pattern collapse is avoided while a scatterometry measurement is provided having a greater sensitivity to manufacturing process variation than the in-chip circuit element.